Discriminative Ability of WordNet Senses on the Task of Detecting Lexical Functions in Spanish Verb Noun Collocations

Collocations, or restricted lexical co-occurrence, are a difficult issue in natural language processing because their semantics cannot be derived from the semantics of their constituents. Therefore, such verb-noun combinations as “take a break,” “catch a bus,” “have lunch” can be interpreted incorrectly by automatic semantic analysis. Since collocations are combinations frequently used in texts, errors in their analysis cannot be ignored. The quality of analysis of collocations can be improved if they are annotated with lexical functions that represent semantic classes of collocations. In this work, we study how WordNet senses viewed as sets of hypernyms can distinguish lexical functions of Spanish verb-noun collocations in experiments with supervised machine learning methods. We show that WordNet senses discriminate lexical functions to different degrees depending on the function, and this phenomenon can be used to evaluate the quality of word sense definitions as well as to measure similarity of various senses of a word and the correlation between word senses and lexical functions.